Download Linguistic Models - Geert Booij`s Page

Linguistic Models
The publications in this series tackle crucial
problems, both empirical and conceptual, within the
context of progressive research programs. In
particular Linguistic Models will address the
development of formal methods in the study of
language with special reference to the interaction
of grammatical components.
Series Editors:
Teun Hoekstra
Harry van der Hulst
Other books in this series:
\
2
3
4
5
6
1
8
9
10
11
12
13
14
15
Michael Moortgat, Harry van der Hulst and Teun Hoekstra (eds)
The Scope of Lexical Rules
Harry van der Hulst and Norval Smith (eds)
The Structure of Phonological Representations. Part I
Harry van der Hulst and Norval Smith (eds)
The Structure of Phonological Representations. Part 11
Gerald Gazdar, Ewan Klein and Geoffrey K. Pullum (eds)
Order, Concord and Constituency
W. de Geest and Y. Putseys (eds)
Sentential Complementation
Teun Hoekstra
Transitivity. Grammatical Relations in Government-Binding Theory
Harry van der Hulst and Norval Smith (eds)
Advances in Nonlinear Phonology
Harry van der Hulst
Syllable Structure and Stress in Dutch
Hans Bennis
Gaps and Dummies
Ian G. Roberts
The Representation of Implicit and Dethematized Subjects
Harry van der Hulst and Norval Smith (eds)
Autosegmental Studies on Pitch Accent
Harry van der Hulst and Norval Smith (eds)
Features, Segmental Structure and Harmony Processes (Part II)
D. Jaspers, W. Klooster, Y. Putseys and P. Seuren (eds)
Sentential Complementation and the Lexicon
René Kager
A Metrical Theory of Stress and Destressing in English and Ducth
I.M. Roca (ed.)
Logical Issues in Language Acquisition
Thematic Structure
Its Role in Grammar
I.M. Roca (ed.)
FORIS PUBLICATIONS
Berlin • New York 1992
Foris Publications Berlin • New York (formerly Foris Publications, Dordrecht) is a Division
of Walter de Gruyter & Co., Berlin.
® Printed on acid-free paper which falls within in the guidelines of de ANSI to ensure permanence and durability.
Library of Congress Cataloging in Publication Data
Thematic structure : its role in grammar / I.M. Roca, ed.
p. cm. - (Linguistic models ; 16)
Includes bibliographical references and indexes.
ISBN 3-11-013406-3 (alk. paper)
1. Grammar, Comparative and general - Topic and comment. I. Roca, Iggy. II. Series.
P298.T47 1992
91-46344
415-dc20
Cl P
Die Deutsche Bibliothek Cataloging in Publication Data
Thematic structure : its role in grammar / I.M. Roca, (ed.) •
Berlin ; New York : Foris Publ., 1992
(Linguistic models ; 16)
ISBN 3-11-013406-3
NE: Roca, Iggy M. [Hrsg.]; GT
• Copyright 1992 by Walter de Gruyter & Co., D-1000 Berlin 30
All rights reserved, including those of translation into foreign languages. No part of this book
may be reproduced or transmitted in any form or by any means, electronic or mechanical,
including photocopy, recording, or any information storage and retrieval system, without
permission in writing from the publisher.
Printing: ICG Printing, Dordrecht
Printed in The Netherlands.
Table of Contents
List of Contributors
xi
Foreword
xv
Martin Atkinson
Introduction
1
Mark C. Baker
Thematic conditions on syntactic structures:
evidence from locative applicatives
1. Background: benefactive and instrumental applicatives
2. The problem of locative applicatives
2.1. Chichewa
2.2. Kinyarwanda and Sesotho
3. The role of the applied affix
4. Contrast 1: Chichewa vs. Sesotho
5. Contrast 2: Kinyarwanda vs. Chichewa/Sesotho
6. Chichewa benefactive applicatives revisited
7. Implications
Notes
References
23
25
29
29
30
32
33
38
40
43
44
46
Geert Booij
Morphology, semantics and argument structure
1. Introduction
2. Lexical-conceptual structure
3. Morphological operations on LCS
4. Verbal prefixation and LCS
5. Middle verbs
6. Conclusions
Notes
References
47
47
48
51
54
58
62
62
63
vi
I. M. Roca
Anna-Maria Di Sciullo
Deverbal compounds and the external argument
1. English déverbal compounds
2. Italian déverbal compounds
3. Italian and English
4. English déverbal compounds revisited
5. Summary
Notes
References
Carlos Gussenhoven
Sentence accents and argument structure
1. Bresnan's account: the Nuclear Stress Rule
2. The generalization
2.1 Difficulty of perceiving nonfinal accents
2.2. Two PP's in one argument
3. Exploiting the generalization
3.1 Multiword predicates
3.1.1. Natural predicates
3.1.2. Newman's pair
3.1.3. Composition
3.2. Embedded clauses
3.2.1. "Raising" vs. "Equi-NP" constructions
3.2.2. PP as predicate
4. Conclusion
Appendix
Notes
References
Ken
The
0.
1.
2.
3.
4.
5.
6.
7.
8.
9.
Hale and Samuel Jay Keyser
syntactic character of thematic structure
Introduction
The double object construction
Conflation as incorporation
Other patterns of conflation
Lexical representations
D-structure representations
Problems of method and the determination of
Lexical Relational Structures
Conflations and the English middle construction
Conflation from the inner subject position
The relational structures of location and locatum verbs .
65
65
67
71
73
76
76
78
79
80
83
87
88
88
88
89
89
92
95
95
98
100
101
104
105
107
107
108
Ill
115
118
122
124
126
132
135
Table of Contents
vu
10. Functional categories in the lexicon
11. Some final remarks
Notes
References
136
138
140
141
Teun Hoekstra
Aspect and Theta Theory
1. The problem
2. Resultatives
2.1. Small clauses
2.2. Resultatives
2.3. Conclusion
3. Licensing
3.1. The theory of modification
3.2. Aspect and Aktionsart
3.3. Licensing Resultatives
3.4. Accomplishments
4. Morphologically complex accomplishments
5. Morphologically simplex verbs
6. Conclusion
Notes
References
145
145
146
146
150
153
153
154
156
161
162
165
169
171
172
173
Richard Hudson
Raising in syntax, semantics and cognition
1. Introduction and overview
2. Raising patterns in syntax
3. Raising patterns in semantics
4. The function of raising
5. Raising patterns in cognition
6. Conclusion
Notes
References
175
175
177
182
189
192
196
197
197
Mary Laughren
Secondary predication as a diagnostic of underlying structure in
Pama-Nyungan languages
1. Part-whole constructions in Warlpiri
2. Stative predicates in Warlpiri
3. Secondary predicates in Arrernte-type languages
4. Lexical structure of verbs
5. "Perception" predicates
6. Conclusion .
199
200
204
207
214
221
237
viii
LM. Roca
Notes
References
240
245
Beth Levin and Malka Rappaport Hovav
The lexical semantics of verbs of motion:
the perspective from unaccusativity
1. Verbs of motion: a problem for the unaccusative hypothesis? . . . .
2. Three classes of verbs of motion
3. Correlations of meaning components with unaccusativity
3.1. Arrive verbs
3.2. Manner of motion verbs
3.3. Run verbs with directional phrases
3.4. Summary
4. Implications for lexical semantic representation
5. Conclusion
Notes
References
247
251
252
253
253
254
258
260
260
265
265
267
M. Rita Manyni
The projection principle(s): a reexamination
1. A review
2. The projection principle
3. The projection principle and the extended projection principle
References
John Charles Smith
Circumstantial complements and direct objects in the Romance
languages: configuration, Case, and thematic structure
1. A traditional view of the problem
2. Some tests for "Objecthood"
2.1. Commutation tests
2.2. Passive constructions
2.3. Causative constructions
2.4. Auxiliary selection
2.5. Extraction
2.6. Cliticization
3. A configurational account
4. Case and Theta-Role assignment
5. Some further data
6. Conclusion
Notes
References
271
272
279
. . . 285
291
293
293
295
295
297
298
299
301
302
303
306
310
313
315
315
Table of Contents
ix
Author index
317
Subject index
321
Morphology, semantics and
argument structure*
Geert Booij
Free University of Amsterdam
I . INTRODUCTION
Traditionally, morphology deals primarily with the internal structure of complex
words, the internal syntax of words. However, morphological theory also has to
account for the effects of morphological processes on the syntactic valency of
the words they create, the so-called external syntax of words.
The minimal level of abstraction that is required to account for the syntactic
effects of morphology is the level of argument structure. For instance, the wellknown correspondence between verbs and déverbal nominalizations can only be
expressed at the level of argument structure (Amritavalli 1980, Rappaport 1983).
Consider, for instance, the data in (1):
(1)
a. John refuses the offer
b. *John refusal the offer
c. John's refusal of the offer/the refusal of the offer by John
In (1), the verb refuse and the déverbal noun refusal have the same argument
structure, with two arguments, usually called Agent and Theme. The differences
in the syntactic realization of these arguments are determined by independent
syntactic principles: (Ib) is ungrammatical because the nominal head refusal
does not assign case to postnominal complements (and hence they have to be
governed by prepositions), and requires prenominal complements to bear the
genitive case ('s). These requirements are fulfilled in (le). Note that in this case
of déverbal nominalization the argument structure of the input words is not
different from that of the output words. This phenomenon is referred to as
'inheritance' of argument structure.
There are also morphological processes in which the argument structure, and
hence the syntactic valency of input words and output words differ. In a by now
classic paper, Williams (1981) argued that the effects of morphological
processes on syntactic valency should be described in terms of operations on
argument structure (i.e. 0-grids). The 0-grid of each verb consists of one or
more 0-roles. At most one of these is external (indicated by underlining), i.e.
has to be realized as subject, outside the VP of which the verb is the head. The
other arguments are internal. Following Marantz (1984) and Levin and
48
Geert Booij
Rappaport (1986) we distinguish two kinds of internal arguments, direct and
indirect ones. A direct argument is realized as a bare NP whereas an indirect
argument is realized as an NP with some oblique case (eg. dative) or as a PP.
Williams claimed that morphological operations on argument structure are
restricted to the following ones:
(a)
E(X): 'Externalize the internal argument, and make the original external
argument internal' (for example -aWe-suffixation, in which according to
Williams the Theme of the verb is externalized, and the Agent
internalized: read Ag. Th -» readable Th. Ag). That is, X has the value
'Th' here);
(b)
E(0): 'Make the external argument internal' (e.g. see AJ; Th -» seen Ag
Th);
(c)
I(X): 'Internalize the external argument, and add a new external
argument' (e.g. random Th -» randomize Ag Th).
Williams' proposal entails that morphological rules can refer to the content of
6-roles. For instance, the characterization of -/'ze-suffixation refers to the 6-role
'Theme'. In the more recent literature on this issue we also find the more
restrictive claim that morphological rules do operate on argument structure, but
do not refer to the content of 6-roles. I.e., morphological rules can only refer to
notions such as 'external argument' and 'internal argument' (Levin and
Rappaport 1986, Pagan 1988).
The hypotheses that I will defend in this paper are the following:
(i)
(ii)
The lexical representation of verbs contains, apart from phonological
information, at least the following two, related, levels: Lexical-Conceptual Structure (LCS) and Predicate-Argument Structure (PAS), where
PAS is a projection of LCS;
Morphological rules may either apply at the level of LCS (and hence
may also indirectly affect PAS), or the level of PAS.
If these hypotheses are correct, they refute Jackendoff's (1987: 379) claim that
there is no motivation for a level of PAS in addition to the level of LCS.
The paper also lends support to the claim that 6-roles do not play a role in
morphology, contrary to Williams' (1981) position.
2. LEXICAL-CONCEPTUAL STRUCTURE
In the spirit of Hale and Keyser (1986), Zubizarreta (1987) and Rappaport,
Levin and Laughren (1988), we distinguish between two levels in the lexical
representations of verbs: LCS and PAS. For instance, the lexical representation
of the transitive verb to eat can be represented as follows:
Morphology, semantics and argument structure
(2)
49
eat
LCS: [x EAT y]ACTION
PAS: V; X, y
LCS specifies the semantic structure of the verb to eat: it expresses a specific
action, which is indicated by the concept EAT, an action in which two
participants are involved, indicated by the ordered pair of variables x and y
(traditionally referred to as the logical subject and the logical object respectively). The concept EAT may be decomposable into smaller conceptual units,
among which the concept ACTION. The minimal assumption that we make here
is that EAT is categorized as a instantiation of the category of ACTIONs.
The level of PAS not only specifies the number of arguments that have to be
realized syntactically, but also contains a minimal form of syntactic annotation:
it specifies that the first argument is the external argument, to be realized as
subject, and that the other argument is the direct internal argument. Thus, the
level of PAS can be qualified as containing "those aspects of lexical meaning
that are grammatically relevant" (Zubizarreta 1987) or as "the part of conceptual
structure that is visible to the syntax" (Jackendoff 1987: 405). It is PAS that
mediates the mapping of LCS onto syntactic structure. At the level of LCS we
specify that to eat is an action with two participants. Since eat expresses an action, the logical subject, i.e. the eater, predictably belongs to the class of agents,
and the logical object, that which is eaten, belongs to the class of patients.
Labels such as ACTION are motivated in Jackendoff (1983).' They are
necessary for a proper account of the aspectual properties of sentences. We also
need them for making generalizations concerning the relation between LCS and
PAS. For instance, it is a well established universal rule (cf. Anderson 1977)
that "in all languages, agent theta roles are external, and patient/theme theta
roles are internal when a verb has both" (Baker 1988: 37). Note that we can
make this generalization without assigning labels for specific 6-roles at the level
of PAS: the relevant generalization can be made by referring to the level of
LCS, where the conceptual unit ACTION is available. As Jackendoff (1987:
378) puts it: "thematic relations are to be reduced to structural configurations
in conceptual structure; the names for them are just convenient mnemonics for
particularly prominent configurations".
Furthermore, the syntactic realization of the two arguments as bare NPs is the
normal realization of external and direct internal arguments, and hence also
predictable.
It is a matter of discussion whether we also need a level of syntactic
subcategorization in addition to LCS and PAS, given the fact that PAS already
contains a minimal form of syntactic information. I will assume here that such
a third level is superfluous, probably with the exception of lexical idioms such
as to look for, to look after, etc., where the selection of the proper preposition
cannot be predicted on semantic grounds.
50
Geert Booij
As already pointed out above, I will support the claim (made by e.g.
Zubizarreta (1987), Rappaport et al (1988) and Di Sciullo (1988)) that 6-role
labels do not play a role in morphology. Their claim is that the only information
concerning arguments at PAS is whether they are external or internal, and
whether they are direct or indirect. Note that this position does not affect the
possibility of making the generalization that in a PAS with an Agent and a
Theme, the Agent is the external argument: as pointed out above, this can also
be expressed directly, without the 0-role labels as intermediaries: the logical
subject of an ACTION always projects as an external argument.
A first, non-morphological illustration of the different roles of LCS and PAS
can be given by comparing the transitive verb to eat with its intransitive
counterpart. At the level of LCS, both have a logical object. However, in the
intransitive use of to eat, the value of the object variable has been fixed as
'food', whereas in the case of the transitive verb the object still is a variable at
the level of PAS. Hence, although there is nothing wrong with the sentence
John eats paper, the sentence John eats can only be interpreted as stating that
John eats food. Therefore, the lexical representation of intransitive to eat is as
follows:
(3)
LCS: x EAT y, y - FOOD
PAS: V; x.
Since the logical object is a constant at the level of LCS (the y-variable is
bound by the conceptual unit FOOD), it is no longer a variable, and hence it
cannot be projected on the level of PAS, and hence this verb is intransitive.
Another illustration of the differences between LCS and PAS is provided by
the lexical representations of the transitive and the ergative to break:
(4)
transitive
LCS:[x CAUSE [(y BREAK)]PROCESS]ACnoN
PAS: V; x, y
ergative
[y BREAK]PROCESS
V; y
These representations express that the differences between the two verbs to
break at the level of PAS (two arguments versus one argument) reflect a more
specific semantic difference at the level of LCS: the Agent of the transitive verb
expresses the causer of the process BREAK (where BREAK may be further
decomposed into COME TO BE BROKEN). Moreover, PAS also expresses that
the only argument of the ergative to break is an internal argument which is a
sister of the V at the level of deep structure, the so-called ergative hypothesis
(cf. Hoekstra 1984 for a survey of the arguments for this interpretation of verbs
such as to break). The syntactic property of ergativity for the superficially
intransitive verbs such as to break follows from the generalization that verbs
that indicate a change of state are normally ergative.
Morphology, semantics and argument structure
51
3. MORPHOLOGICAL OPERATIONS ON LCS
Given the assumption outlined above that the lexical representations of verbs
contains at least the levels of LCS and PAS, the question now arises whether
morphological rules apply to LCS or to PAS. Below, I will show that both kinds
of morphological processes have to be allowed for, i.e. some rules apply at the
level of LCS, whereas other rules apply at the level of PAS.
Knopper (1984) has shown that the Dutch déverbal suffix -sel has to be
qualified as a suffix that creates names that refer to the internal argument of the
verb. This is illustrated in (5): in (5a) we find object names derived from
transitive verbs, (5b) shows examples of -sel-nouns derived from ergative verbs
which have only one argument. The criteria for the ergativity of Dutch verbs are
taken from Hoekstra (1984). One of them is that ergative verbs select the
auxiliary verb zijn 'to be' instead of hebben 'to have' which is selected by
intransitive and transitive verbs. The examples are from Knopper (1984):
(5)
a.
b.
tik-sel 'typing'
from
bak-sel 'baking'
uitwerp-sel 'excrement'
aanslib-sel 'deposit'
verschijn-sel 'phenomenon'
bezink-sel 'sediment'
tik 'to type-write'
bak 'to bake'
uitwerp 'to excrete'
aanslib 'to form a deposit'
verschijn 'to appear'
bezink 'to settle'
Crucially, intransitive, non-ergative verbs do no allow for -^^/-suffixation, and
hence the relevant generalization can only be stated at the level of PAS: at the
level of LCS the arguments of intransitive verbs have the same status as those
of ergative verbs, that of 'logical subject'. It is at the level of PAS that the
syntactic difference between them (external versus internal argument) is stated,
and therefore, the relevant word formation rule has to apply at PAS.
A second example of a word formation process that is to be accounted for at
the level of PAS is the formation of déverbal -er-nouns in English. Levin and
Rappaport (1988) have argued that the relevant generalization is that déverbal
-er creates nouns that refer to the external argument of verbs. Thus, it is
predicted that -er does not attach to ergative verbs, and that, for instance, the
following nouns are ill-formed (Levin and Rappaport 1988: 1075):
(6)
*disappearer, *happener, *dier, *laster, *ender, *exister, *occurrer,
*collapser
However, these analyses should not be taken to imply that all morphological
processes apply at the level of PAS. Below, I will discuss a number of morphological processes that must be accounted for in terms of an operation on the
LCS of verbs, with concomitant effects on PAS, and hence on syntactic valency.
Geert Booij
52
The first case study is that of déverbal nouns in -er in Dutch, which look
similar to those of English. The suffix -er creates so-called subject names, as
illustrated in (7). The base verbs are given in their citation form, the infinitive,
which consists of stem + -en:
(7)
spel-er 'player'
fiets-er 'cyclist'
prat-er 'talker'
<spelen 'to play'
<fietsen 'to cycle'
<praten 'to talk'
As pointed out in Booij (1986), déverbal -er-suffixation not only creates agent
nouns, but also nouns with non-agentive interpretation, for instance:
(8)
lijd-er 'sufferer'
(alles)kunn-er 'lit. all-canner,
who can do everything'
hebb-er 'lit. haver, vulture'
durv-er 'lit. darer, daredevil'
<lijden 'to suffer'
<kunnen 'can'
<hebben 'to have'
<durven 'to dare'
This already shows that this word formation process should not refer to theta
role labels such as Agent, but only to the notion 'subject'. The question now is,
whether this notion 'subject' refers to PAS and hence to the notion 'external
argument', or to LCS, i.e. to the notion 'logical subject'. Hoekstra (1984) has
argued that the notion 'external argument' is crucial, because -er cannot attach
to ergative verbs. That is, Dutch -er is claimed to behave like its English
counterpart. However, Booij (1986) argued that this claim is empirically
incorrect, because there is a substantial number of -er-nouns derived from
ergative verbs. Moreover, this class does not form a closed set, and can be
extended. Examples of this class of nouns derived from ergative verbs are given
in (9):
(9)
breker 'wave that breaks'
stijger 'riser'
daler 'descender'
groeier 'grower'
zinker 'lit. sinker, underwater
main'
blijver 'stayer'
beginner 'beginner'
ontsnapper 'escaper'
binnenkomer 'enterer'
afzwaaier 'who leaves military
service'
binnendringer 'invader'
<breken 'to break'
<stijgen 'to rise'
<dalen 'to descend'
<groeien 'to grow'
<zinken 'to sink'
<blijven 'to stay'
<beginnen 'to begin'
<ontsnappen 'to escape'
<binnenkomen 'to enter'
<afzwaaien 'to leave military service'
<binnendringen 'to invade'
Morphology, semantics and argument structure
uitvaller 'drop out'
meevaller 'piece of good luck'
afvaller 'drop out'
invaller 'stand in'
insluiper 'sneak-thief'
53
<uitvallen 'to drop out'
<meevallen 'turn out to be better
than expected'
<afvallen 'to drop out'
<invallen 'to stand in'
<insluipen 'to sneak in'
Clearly then, the only valid generalization is that Dutch deverbal -er creates
names for logical subjects, that is, subjects at the level of LCS. It is true that
certain ergative verbs do not have a related noun in -er. For instance, the Dutch
counterpart of English *appearer, blijker, does not exist, and this was one of the
facts that Hoekstra (1984) adduced in favour of the hypothesis that -er does not
attach to ergative verbs. Note, however, that the verb schijnen 'to seem' also
does not allow for -er-suffixation, although it is not ergative. Some independent
explanation should therefore be sought for the illformedness of *blijker and
*schijner. Note, for instance, that -er creates names for entities, whereas the
subject of the verbs blijken and schijnen is a proposition. This semantic 'clash'
may therefore explain the illformedness of these words in -er.
Another morphological process that, I would like to argue, applies at LCS is
Dutch déverbal -toar-suffixation, comparable to -aWe-suffixation in English.
Recall that Williams (1981) qualified -aWe-suffixation as an operation by which
the original internal argument becomes external, and vice versa. For instance,
read Ag Th becomes readable Th Ag. However, in Dutch -foar-suffixation the
Agent-argument is clearly lost:2
(10)
a.
b.
Dit verschijnsel is verklaarbaar (*door mij)
'This phenomenon is explainable (*by me)'
Dit boek is goed leesbaar (*door mij)
'This book is well readable (*by me)'
The loss of the the Agent argument also manifests itself in the fact that the
Agent argument no longer functions as an implicit argument that can play a role
in control, and in this respect -baar/-able suffixation differs from the passive:
(11)
a. *This book is readable to learn the facts of life
b. This book was read to learn the facts of life
The semantic characteristic of -fcaar-sufftxation is that it creates adjectives
which express a potentiality property of the Theme-argument with respect to the
action expressed by the base verb. The agent phrase (by-PP) that occurs with
passives, is not permitted here. Instead, it is possible to use the PP voor mij 'for
me': Dit boek is goed leesbaar voor mij 'This book is well readable for me'.
Note that voor mij is not an agentive phrase, but is rather used in combination
54
Geert Booij
with expressions that denote properties, as in Die jurk is leuk voor mij 'That
dress is nice for me', with the underived adjective leuk 'nice'.
Hence, the LCS of -fcaar-adjectives can be defined as follows:
(12)
The property of y such that, for some xMb, it is possible that
x PREDICATE y (where x PREDICATE y is the LCS of the input verb)
Following Pagan (1988), we consider xKb as the expression of the generic
interpretation of a variable, in this case of -boar-adjectives, and formally as an
operator that binds the variable x. Hence, it will not project to the level of PAS.
This accounts for the impossibility of the agentive door-phrase to cooccur with
-baar-adjectives. The only remaining variable is the logical object variable,
which projects as external argument on PAS. We assume that when LCS
contains only one logical variable, this variable will in the unmarked case
project as external argument (i.e. ergative verbs form an exceptional class in this
respect).
If we wanted to describe deverbal -baar-suffixation as an operation on PAS
instead of an operation on LCS, we would have to describe it as follows: delete
the external argument, and make the internal argument external. However, this
description does not express the fact that there is still an understood generic
agentive subject for such adjectives. Moreover, this word formation rule has to
apply to LCS anyway, because the 'potentiality' interpretation of such adjectives
can only be expressed at that level. Therefore, the most straightforward and
generalizing account of -eoar-suffixation is the one in terms of an operation on
the LCS of its base verbs, with predictable consequences for the level of PAS.
In the next section I will present another case of déverbal affixation which
can only be stated insightfully and in a generalizing fashion in terms of an
operation on LCS.
4. VERBAL PREFIXATION AND LCS
Dutch has a number of déverbal prefixes, among which be-. As shown in (13),
the prefix be- attaches to verbs with different types of argument structure.
Nevertheless, the output verbs always have the same PAS: x., y. That is, they are
obligatorily transitive verbs. A uniform output makes it impossible to state the
rule as an operation on PAS because we would then predict different output
argument structures correlating with different input argument structures.
Therefore, the rule must be stated as an operation on LCS.
Morphology, semantics and argument structure
(13)
a.
b.
c.
d.
type of input verb:
external arg. only:
lopen
'to walk'
klimmen
'to climb'
with indirect int. arg. :
twijfelen aan iets
'doubt about something'
vechten met iemand
'fight with somebody'
slapen met iemand
'sleep with somebody'
with direct int. arg.:
schilderen
'to paint'
plakken
'to glue'
with dir. and indir. arg:
iets roven van iemand
'rob something from
somebody'
iets planten in iets
'plant something in something'
55
output verb:
iets belopen
'walk on something'
iets beklimmen
'to climb on something'
iets betwijfelen
'doubt something'
iets bevechten
'fight something'
iemand beslapen
'lit. sleep on somebody'
iets beschilderen
'to put paint on something'
iets beplakken
'to put glue on something'
iemand beroven van iets
'to rob somebody by taking away
something'
iets beplanten met iets
'plant something with something'
The same morphological process of fee-prefixation occurs in the related
Germanic languages German (cf. Günther 1986, Wunderlich 1987) and Frisian
(Veenstra 1988). Both Günther and Wunderlich primarily deal with the subclass
of fee-prefixation exemplified in (13d). A similar alternation is found in English
(cf. Rappaport et al. 1988), but without overt morphological marking:
(14)
a.
b.
to plant trees in the garden/to plant the garden with trees
to load hay onto the wagon/to load the wagon with hay
Günther's (1986) and Wunderlich's (1987) analyses of the German equivalents
of the cases (13d) are essentially as follows: sentences with underived verbs and
sentences with the derived fee-counterparts are paraphrases of each other. Beprefixation only affects the syntactic realization of the different 0-roles: with
underived verbs, the Location is expressed as an indirect argument (i.e. as a PP
with a specific preposition), whereas with the fee-verbs the Location-role is the
direct argument, and the original direct internal argument becomes the indirect
one. This is illustrated in (15):
56
(15)
Geert Booij
a.
b.
Er giesst [Wasser]dir [auf die Blumen]ind
'He pours water on the flowers'
Er begiesst [die Blumen]dir [mit Wasser]^
'He covers the flowers with water'
In this view then, te-prefixation is not seen as a process that primarily affects
LCS, but as an operation affecting PAS because it only affects the syntactic
realization of what is assumed to be the common meaning of both the simplex
verb and the derived fee-verb.3 Therefore, both Günther and Wunderlich consider
this process as comparable to Passive formation.
Two objections must be raised to the kind of analysis proposed by Günther
and Wunderlich. First, it is not expressed in this approach that the object of the
derived verb is always an object that is completely affected by the action
expressed by the verb (as a matter of fact, a very traditional observation
concerning he-verbs). For instance, the expression de tuin met rozen beplanten
'to plant the garden with roses' implies that the garden is completely covered
with roses, whereas this is not necessarily the case when the expression rozen
in de tuin planten 'to plant roses in the garden' is used. In this connection the
semantic opposition particularistic : holistic is also used. According to
Wunderlich (1987: 303) it is not always the case that the object is completely
affected. For instance, sentence (15b) does not necessarily mean that the flowers
are completely covered with water. Although this may be correct, even in that
case the phrasing with the fee-verb suggests that somehow the flowers are
completely affected, although the water may not completely cover the flowers.
The second point to be made is that by giving a special rule of fee-prefixation
for the verbs of type (13d), we do not express the generalization that, whatever
the argument structure of the input verb may be, the argument structure of the
output verb is always the same, as shown in (13).
Therefore, déverbal fee-prefixation should be seen as a morphological process
that creates verbs that express an action that completely affects the object of that
action. Since affected objects always project as direct internal arguments on the
level of PAS, the PAS, and hence the syntactic transitivity of all déverbal heverbs is completely predictable. The LCS of déverbal fee-verbs is expressed in
(16):
(16)
x completely affects y by executing the ACTION of P-ing, where P is
the Predicate of the input verb.
This LCS generalizes across all kinds of input verbs, because the predicate itself
is the only element taken over from the LCS of the input verb, not the
arguments; the latter are created by (16).
This analysis implies that even in the case of transitive input verbs, the
arguments of the output verb are created. This position is corroborated by the
Morphology, semantics and argument structure
57
fact that transitive underived and derived verbs impose different selection
restrictions on their arguments. For instance, the verb schilderen 'to paint' can
cooccur with an effected object (as in een landschap schilderen 'to paint a
landscape') whereas the object of beschilderen 'to cover with paint' is always
an affected object. The same difference is found in the pak schrijven 'to
write '/beschrijven 'to write on' (compare een briefschrijven 'to write a letter'
to een vel papier beschrijven 'to cover a piece of paper with writing'.
Consequently, in such cases the direct argument of the underived verb cannot
be the same as that of the derived counterpart.
As shown in Booij and Van Haaften (1988) exactly the same holds for
déverbal ver-prefixation in Dutch: the input verbs of ver-prefixation differ in
argument structure, but the ver-verbs are always obligatorily transitive verbs.
Therefore, ver-prefixation must also be described as a case of creation of
argument structure. The common semantic aspect of ver-verbs is that the action
referred to in the base word is directed toward the object in such a way that the
object is really affected by the action. Similar word formation rules are those
that prefix om-, door- or over- to a verb: they always create verbs with a direct
internal argument, whatever the argument structure of the input verb:
(17)
a.
b.
c.
d.
type of input verb
ext. arg. only
schijnen 'shine'
woekeren 'grow rank'
direct int. arg.
zien 'see'
ind. int. argument
denken aan iets
'think of something'
direct + ind. argument
een gat in iets boren
'drill a hole in something'
iets in iets wikkelen
'wrap something in something'
output verb
omschijnen 'shine around'
doorschijnen 'shine through'
overwoekeren 'overgrow'
doorzien 'see through'
overzien 'look over'
iets overdenken
'think about something'
iets doordenken
'grasp something completely'
iets doorboren
'drill through something'
iets omwikkelen
'wrap something completely'
In (13d) we saw two pairs of verbs in which both the input verb and the output
verb seem to have three arguments. Therefore, the question arises how we can
deal with such cases, which suggest that the effect of fee-prefixation has to do
with PAS rather than with LCS. However, it should be observed that in expressions like de tuin met rozen beplanten 'to plant the garden with roses' it is
58
Geert Booij
always possible to omit the met-phrase, as is also observed for German by
Günther (1986: 193). That is, we suggest that the met-phrase is an optional
adjunct, rather than an argument. The adjunct status of such mef-phrases is supported by the fact that they can be separated from the direct object by en wet:
(18)
a. *Hij beplant, en wel de tuin
'He plants, namely the garden'
b. Hij beplant de tuin, en wel met rozen
'He plants the garden, namely with roses'
The upshot of this analysis is that there is no formal similarity between beprefixation and the formation of passive verbs, as was suggested by Günther and
Wunderlich: be-prefixation is an operation at the level of LCS, with concomitant
changes at the level of PAS, whereas Verbal Passive formation is an example
of a morphological operation that directly affects PAS in that it suppresses the
syntactic realization of the external argument. As noted frequently in the
literature, in passives the Agent argument has the status of an implicit argument,
since it can function as a controller in passive sentences (Roeper 1987):
(19)
a. The boat was sunk to collect the insurance money
b. *The boat sank to collect the insurance money
In (19b), the to-infinitive construction cannot be used because for ergative verbs
there is no implicit argument that can function as a controller.
We summarize our conclusions so far as follows: morphological rules can
operate on two different levels of the lexical representation of verbs: LCS and
PAS. When they apply at LCS, this may also affect PAS, because PAS is a
projection of LCS. We have seen the following types of direct operation on
PAS: (i) binding of one of the arguments (English -er binds the external
argument of verbs, Dutch -sel binds the internal argument); (ii) suppression of
an argument at PAS (in verbal passives, the external argument of the base verb
is not realized syntactically, but can function as controller for the PRO-subject
of to-infinitives). Note also that we were able to account for the differences
between English and Dutch déverbal -er-suffixation by assuming that the first
process applies at PAS, and the second one at LCS. In the next section, we will
see that the distinction between PAS and LCS is also crucial for a proper
characterization of the differences between middle verbs and passive verbs.
5. MIDDLE VERBS
English has a productive rule of middle verb formation. As Pagan (1988) argues
convincingly, middle verbs must be considered to be derived by a lexical rule,
Morphology, semantics and argument structure
59
not by a syntactic movement rule. Pagan points out that the underlying subject
of the middle verb is interpreted genetically. For instance, according to her,
sentence (20a) can be paraphrased as (20b):
(20)
a.
b.
This book reads easily
People, in general, can read this book easily
Pagan proposed to account for this observation on genericness by characterizing
the rule of English middle formation as follows:
(21)
a.
b.
Assign arb to the external 0-role
Externalize the direct 6-role
Pagan assumes that by assigning the index arb to a role, it is 'lexically
saturated', and therefore it will not be realized syntactically. Clearly, then,
middle formation is seen by Pagan as an operation that directly affects the level
of PAS.
When we apply (21) to e.g. the verb to read, we get the following PAS for
the middle verb to read:
(22)
read, x^, y(\ - Agent, y - Theme)
As far as I can see, there is some unclarity here in having two external 6-roles.
Let us therefore assume that the saturation of the Agent role is expressed at the
level of LCS, parallel to the representation of the implicit object of the
intransitive to eat:
(23)
LCS: for some xub, x READ y
PAS: V, y.
Due to this operation on LCS, middle verbs are agentless and predictably
intransitive, since there is only one free variable that can project to PAS. The
fact that the original agent variable is bound, can also be observed in the
ungrammaticality of sentences with middle verbs and agentive i>y-phrases:
(24)
*This book reads well by John
However, even then the semantic differences between middles and passives are
not completely expressed. Pagan points out that "middles [...] are not used to
report events, but to attribute a specific property to some object" (p. 200). The
static nature of middle predicates excludes progressive forms as witnessed by
the ill-formedness of *This book is reading easily. This static nature of middle
predicates is not expressed by (23). It thus appears that a more adequate
60
Geen Booij
semantic characterization of middle formation is called for, in terms of an
operation on LCS from which the proper PAS can be projected. I propose the
following LCS for the middle verb read that also expresses the potentiality
aspect of the semantic interpretation:
(25)
the property of y such that, for xub, it is possible that x READ y
In this LCS, the variable x is bound by the operator x^ and hence only the
variable y can be projected on the level of PAS, as the external argument.
In Dutch, middle formation not only applies to transitive verbs, but also to
verbs with an optional indirect internal argument (e.g. op een paard rijden 'to
ride on a horse') and to intransitive verbs such as zingen 'to sing':
(26)
Dat paard rijdt lekker
That horse rides nicely
'On that horse, one can ride nicely'
(27)
Het zingt hier gemakkelijk
It sings here easily
'One can sing here easily'
In (27) there is no external argument for the middle verb zingen, and the subject
position is filled with the expletive pronoun het 'it*.
Thus, the following general semantic characterization of Dutch middle verbs
can be given:
(28)
the property (of y) such that for xub it is possible that
[x PREDICATE (y)]ACnoN
where [x PREDICATE (y)] stands for the LCS of the input verb. The optional
y-variable corresponds to the direct or indirect internal argument of the input
verb. Since LCS (28) contains at most one free variable, the PAS of middle
verbs will either contain one, external, argument, as in (26), or no argument at
all, as in (27).
It should be mentioned here that both passive formation and middle formation
do not apply to ergative verbs, i.e. they only apply to verbs with an external
argument. This is illustrated in (29) where passive formation and middle
formation yield ungrammatical results:
(29)
a. *Er wordt hier vaak gevallen
There is here frequently fallen
'One falls here frequently'
Morphology, semantics and argument structure
61
b. *Het valt hier gauw
It falls here easily
'One can easily fall here'
This may at first sight suggest that middle formation applies at the level of PAS,
since it seems that the relevant class of input verbs is to be defined in terms of
PAS-properties: verbs with an external argument. However, we also saw that
middle formation not only affects the argument structure of the input verbs, but
also implies other, semantic modifications. Hence, we have to assume that
middle formation applies at the level of LCS. The class of verbs that undergo
middle formation can also be defined at that level, namely as the class of
ACTION verbs (this is already presupposed in (28)). Independent evidence for
the correctness of this semantic definition of the class of input verbs is that
psychological verbs such as ergeren 'to annoy', storen 'to disturb' and verbazen
'to amaze', which are non-ergative and have an external argument, cannot be
subject to middle formation, whereas they do allow for passive formation, an
operation on PAS:
(30)
a.
Middle formation
*Jan ergert gemakkelijk
'John annoys easily'
b. Passive formation
Jan wordt gemakkelijk geërgerd
'John is easily annoyed'
This analysis of middle formation also accounts for the difference between
middles and passives with respect to control phenomena. Unlike the implicit
agent of passives, the lexically saturated agent variable of middles is not
projected at PAS, and thus cannot function as controller:
(30)
a. The book was read to solve the problems
b. *This book reads easily to solve the problems
In sum, it seems to me that the differences between middle verbs and passives
can receive an insightful description by making use of the difference between
LCS and PAS: middle verbs are formed by an operation on LCS, with
consequences for PAS, whereas passive verbs only differ from their active
counterparts at the level of PAS: the external argument is present, but
'suppressed', i.e. it is not realized syntactically, although it can still function as
a controller for PRO. This approach is in accordance with Williams' (1987)
claim that control relations are to be stated at the level of argument structure,
where 'implicit arguments' are available. The subjects of the verbal bases of
62
Geert Booij
-aWe-adjectives and middle verbs do not receive the status of implicit
arguments, unlike the subjects of the bases of verbal passives.
6. CONCLUSIONS
Let me now briefly summarize the conclusions of this paper. The most
important point was that morphological operations that affect syntactic valency
should not always be characterized in terms of operations on argument structure,
but often as operations at the level of Lexical Conceptual Structure. This
situation is to be expected given the existence of two levels of lexical
representation, LCS and PAS, which were also motivated for simplex verbs such
as the two verbs to eat and the two verbs to break. Since PAS is a projection
of LCS, morphology may also have effects on PAS due to changes at LCS.
Another conclusion is that the phenomena discussed above do not require that
the arguments at the level of PAS are labeled for a specific 6-role: generalizations that refer to PAS can be expressed in terms of the distinction between
external and internal argument, and that between direct and indirect arguments.
At a more general level we saw the importance of semantics for a proper
characterization of the distributional properties of derived words. In the words
of Pagan (1988: 202):
This discussion [...] brings to light some of the problems encountered when one searches for
syntactic explanations for linguistic phenomena without fully considering semantic alternatives.
We cannot ignore the complex interaction of semantic and syntactic features (hat underly any
given construction. If we give due consideration to all aspects of grammar (syntax, semantics,
and so on) we will stand a better chance of more fully understanding the linguistic phenomena
we set out to explore.
By showing that many morphological operations must be described in terms of
operations at LCS, we established evidence for the crucial role of semantic
generalizations and insights in morphology. What remains to be developed is a
more fully articulated theory of the structure of representations at LCS, and of
the linking or projection rules that relate LCS to PAS.
NOTES
* The research for this paper was carried out as part of research programme Lelt 88/9, Vrije
Universiteit Amsterdam. I would like to thank the audiences at the University of Essex, the Vrije
Universiteit Amsterdam, and the University of Lille for their reactions, and Ton van Haaften,
Rochelle Lieber, the editor, and the anonymous referees for their comments on a previous version
of this paper.
Morphology, semantics and argument structure
63
1. Semantic notions like Activity (- Action) and Process are also indispensable for a correct
characterization of the distribution of small clause complements of verbs (cf. Hoekstra 1988).
2. According to Randall (1982) -oWe-adjectives are similar to -/VMr-adjectives in that they do not
allow for by-phrases.
3. As Wunderlich (1987: 300) observes, such an analysis conflicts with Williams' (1981)
hypothesis that morphology cannot only affect internal arguments.
REFERENCES
Amritavalli, A. 1980. Expressing Cross-categorial Selectional Correspondences. An Alternative to
the X-bar Syntax Approach. Linguistic Analysis 6. 305-343.
Anderson, S.R. 1977. Comments on the Paper by Wasow. In P.W. Culicover, T. Wasow, A.
Akmajian (eds.) Formal Syntax. New York: Academic Press. 361-377.
Baker, M. 1988. Incorporation. Chicago: University of Chicago Press.
Booij, O.E. 1986. Form and Meaning in Morphology: the Case of Dutch 'Agent Nouns'. Linguistics
24. 503-517.
Booij, O.E. and T. van Haaften. 1988. The External Syntax of Derived Words, Evidence from
Dutch. Yearbook of Morphology 1. 29-44.
Di Sciullo, A.-M. 1988. Formal Relations and Argument Structure. Paper given at the 3rd Int.
Morphology Meeting, Krems, July 1988.
Pagan, S.M.B. 1988. The English Middle. Linguistic Inquiry 19. 181-204.
Günther, H. 1986. Wortbildung, Syntax, i«-Verben und das Lexicon. Beiträge zur Geschichte der
deutschen Sprache und Literatur 109. 179-201.
Haie, K. and S.J. Keyser. 1986. Some Transitivity Alternations in English. Lexicon Project Working
Papers 7. Cambridge, Mass.: MIT.
Hoekstra, T. 1984. Transitivity. Dordrecht: Foris.
Hoekstra, T. 1988. Small Clause Results. Lingua 74. 101-139.
Jackendoff, R.S. 1983. Semantics and Cognition. Cambridge, Mass.: MIT Press.
Jackendoff, R.S. 1987. The Status of Thematic Relations in Linguistic Theory. Linguistic Inquiry
18. 369-412.
Knopper, R. 1984. On the Morphology of Ergalive Verbs and the Polyfunctionality Principle. In H.
Bennis and W.U.S. van Lessen Kloeke (eds.) Linguistics in the Netherlands 1984. Dordrecht:
Foris. 119-128.
Levin, B. and M. Rappaport. 1986. The Formation of Adjectival Passives. Linguistic Inquiry 17.
623662.
Levin, B. and M. Rappaport. 1988. Non-event er-nominals: a Probe into Argument Structure.
Linguistics 26. 1067-1084.
Marantz, A. 1984. On the Nature of Grammatical Relations. Cambridge, Mass.: MIT Press.
Randall, J. 1982. Morphology and Language Acquisition. Doctoral dissertation, University of
Amhersl.
Rappaport, M. 1983. On the Nature of Derived Nominals. In M. Levin, M. Rappaport and A.
Zaenen (eds.) Papers in Lexical Functional Grammar. Bloomington, Ind.: IULC. 113-142.
Rappaport, M., B. Levin, and M. Laughren. 1988. Niveaux de représentation lexicale. Lexique 7.
13-32.
Roeper, Th. 1987. Implicit Arguments and the Head-Complement Relation. Linguistic Inquiry 18.
267-310.
Veenstra, D.H. 1988. Oer de Grammatika fan te-ferba. In S. Dyk and G. de Haan (eds.)
Wurdfoarried en Wurdgrammatika. Ljouwert: Fryske Akademy. 137-174.
Williams, E. 1981. Argument Structure and Morphology. The Linguistic Review 1. 81-114.
64
Geert Booij
Williams, E. 1987. Implicit Arguments, the Binding Theory, and Control. Natural Language and
Linguistic Theory 5. 151-180.
Wunderlich, D. 1987. An Investigation of Lexical Composition: the Case of German fe-verbs.
Linguistics 25. 283-331.
Zubizarreta, M-L. 1986. Levels of Representations in the Lexicon and in the Syntax. Dordrecht:
Foris.